Hi everyone,
I want to vectorize multiple matrix-vector products and avoid a for
loop, a little bit like np.linalg.solve does, for instance :
/nDOF = 100//
//M = 4//
//
//mat = np.random.rand(nDOF, M, M)//
//u = np.random.rand(nDOF, M)/
/vecMatInv = np.linalg.solve(mat, u) # => shape (nDOF, M)
/
/for i in range(nDOF)://
// assert np.allclose(vecMatInv[i], np.linalg.solve(mat[i],
u[i])) # => OK
/
This is quite straightforward and also works if u is only a flat vector
of size M (which is great). But this is not that easy for matrix vector
multiplication, and the solution I found yet is to use np.matmul with
some manipulations over the array dimensions :
vecMatMul = np.squeeze(np.matmul(mat, u[..., None]), axis=-1) /# =>
shape (nDOF, M)/
for i in range(nDOF):
assert np.allclose(vecMatMul[i], mat[i] @ u[i]) # => OK
That way we get the same behavior as for np.linalg.solve. But this seems
not very natural, either for np.matmul or np.dot when using arrays of
dimension > 2.
Since np.linalg.solve does this vectorization naturally, I wonder if
there is a way to get the same behavior with matrix-vector
multiplication already in numpy, and why np.matmul does not behave like
np.linalg.solve does ?
Best,
Thibaut
--
*Dr. Thibaut LUNET*
Chair Computational Mathematics,
/Institute of Mathematics (E-10),/
/Office 3.040, tel: +49 40 42878 3428/
*Hamburg University of Technology*
_______________________________________________
NumPy-Discussion mailing list -- numpy-discussion@python.org
To unsubscribe send an email to numpy-discussion-le...@python.org
https://mail.python.org/mailman3/lists/numpy-discussion.python.org/
Member address: arch...@mail-archive.com
